Pushing Ryzen 7000's Infinity Fabric to its limits

[@SwampyF4RT]

This content is why I buy AHOC merch. I know these videos dont get tons of views, but I genuinely appeciate this informative research.

[@ConnorDaly-n7c]

Well you don't want to be wearing a shirt promoting a tacky brand now, do you?

[@jonnyj.]

@@ConnorDaly-n7c Or... maybe he wants to do whatever the fuck he wants to do? As an adult...? Who spends his own money...?

[@SwampyF4RT]

Thanks for always providing real deep dives on tech!

[@Splarkszter]

love this highly technical deep inside. thank you for this kind of content and thank you for all your efforts.

[@Bunster]

thats what she said

[@Basement-Science]

Ryzen 3000 also has error correction in the infinity fabric, IF you enable it. Which, I suspect a lot of motherboards dont let you do. My Asrock X570 Taichi by default has this disabled, and lets you enable error detection and then error correction for infinity fabric. Windows confirms that this works by logging these errors in the event log as either detected or corrected. I dont know about other series, but I dont imagine they would have removed this feature in the 5000 series.

[@Snoop05B]

In my experience wtih 3000 series the error correction started working since some BIOS/Agesa update automatically (previously whole system just crashed), but once it hit error it logged as WHEA but performance was permanently hampered (not recovering unless you reboot).

[@djanto482]

even 5000 has ecc on both core and fclk, the classic whea18 and whea19, but i can't figure how to enable them on 7000 to work with.

[@yup8388]

I can confirm Ryzen 3000 (I own one) has this IF retransmit behavior (but there is no need to enable it, works out of the box) and I'd assume it's been always there since the first Ryzen and IF. As shown in the vid, unstable OC causes dropouts/issues with latency or even crashes. Stuttery mouse and sound artifacts (especially when you're using a PCIE sound card - my case) are just obvious signs of it. I found Latencymon to be an indispensable tool for OC.

[@carsashimi]

When the Ryzen 5000 series came out windows did not report WHEA errors most of the time either, similar to the current situation with 7000 series. There was then either a windows or driver update (not sure which, but I have a feeling that it was windows), after which the WHEA errors started being reported. Hopefully we get that again for 7000 series, although it has been some time since release now, so maybe AMD decided not to expose then this time around.

[@Taraquin83]

I miss whea 18/19 reporting on 7000, getting stable curve optimizer after corecycler is much easier with whea 18-reporting.

[@RobBCactive]

There might be good reasons for silence. Reporting recoverable errors might turn graceful performance degradation into falling of a cliff, as even low re-transmission rates might make error logging over-whelm useful work. Zen4 uses double speed half width IF transmission which may perform better but rely on a low re-transmission rate. Reporting might turn an unnoticed latency spike into a highly visible performance problem. This means an over-clocker sees a drop in performance, but can still correct the setting.

[@N0N0111]

I seen many people push to run 2166 on their DDR5 6200 / 6400 OC. I just gonna keep it at 2133, running it for 3 weeks now no weird instabilities. (btw really cool series with oscilloscope for voltage behaviors for these new platforms!)

[@Directx11cool]

This exactly me, but I run crunch and it shows no error.

[@jrherita]

Is there a reason not to run 2066 with DDR5-6200? 2066 * 3 = 6200 so timing would be more correct (like 2000/6000)

[@Malklaw]

I got muuuuuch better score even missing 1:3 ratio 😄 6200CL28 2200FLCK 1.14 SOC 1.24VDDIO and now testing 1.11 1.21 still same consistent results on 7600x 70k97k70k and 53.3 latency Aida i dont understand this cpu 😂 I saw some barely runs stable at 1.3+ at this settings and thanks to my voltages my cpu temps are much lower now 😂On flck 2100-2133 i got 3k slower results +-

[@drxpsaucx4157]

​@@Malklaw what are your timing settings I'm hoping to get 6200 cl28 on my silicon

[@RobertSmith-lh6hg]

​@@jrherita the *3 thing doesn't work. Best FCLK is the highest FCLk without causing errors. You can detect the silent intermittent errors with linpsack GFlops, AIDA64 latency, and a few other tests. There will be a fclk speed which provides best performance and each step after that ideal one will show regression long before crashes occur. Fabric max stable speed is very temp sensitive.

[@romangruber6685]

This is not a problem. The purpose of error correction is to correct errors. So the task can be finsihed. it is the user task to set stable settings.

[@speedfastestsolo]

Better if they have option to disable or enable it, so if overclock is unstable, easily detectable, after you find stability you can enabling error correction

[@kuyans3889]

I think I'd rather find out that my OC is unstable when the benchmark score comes back low, than by finding out when programs don't open any more or other things get corrupted. Seems like a good system. Maybe there should be an option to turn off this error correction, I'm sure there is some benchmark that benefits.

[@certaindeath7776]

yeah, but i would want my system to tell me in either case. there should be error messages.

[@mortlet5180]

The problem is your benchmarks won't necessarily show worse performance, because there's a trade-off between getting higher scores from the extra 33Mhz, when running at 2133, that first needs to be exceeded by the performance loss incurred from retransmissions, which differs from CPU to CPU and also depends on how hard the IMC has to work, since RAM related work will pull down the IO-die's core voltage (because of lazy engineering and excuses of "the laws of Physics", which aparently prohibits the implementation of any advancements in the field of control systems from the last 50 years!).

[@GewelReal]

@@mortlet5180 but wouldn't you be able to tell by looking on the average of say 100 runs that at some point you stopped scaling meaning you started getting errors?

[@LostScout-8X]

Thank you for the in-depth dive into cause and effect!!

[@DigitalJedi]

This is extremely interesting to see from the Ryzen side of things, would love to see a full writeup charting where there is instability and to what extent. I've documented some of this with RDNA2 and 3 myself, as well as Pascal and Alchemist. I do package engineering for Intel, and prototype testing is a good part of that now that we are nearing architecture refreshes. I deal mostly with the stuff between dies or between them and the socket, but my MS is in architecture design. I can't say what I've seen this sort of behavior on at work, but we have a similar "crawl state" behavior for some MCM devices as well. Of note in the public realm is Sapphire Rapids, where there are links between 4 dies and 4 HBM stacks that can potentially have massive transmission errors. Those are server chips, so slow is better than crashed, so the thing just tries again until it goes through.

[@andytroo]

now this is Actually Hardcore Overclocking - you've attached an oscilloscope to find the error limited performance sweet-spot on the frequency curve. Is there any options to expose these issues to the OS? - it would make sense to say "warning, performance exceeding limitations, errors detected"

[@JJFX-]

Awesome content. Visualizes perfectly what most of us experience. My 7950X is about 1 step above this. Consistently scales in everything up to 2167, even at 6400 MT/s. If there are drop-outs, it's not enough to cancel out the gains. At 2200, I've never thrown an error or but scores, bandwidth and latency take a clear dive. Same behavior you're seeing at 2167. I haven't seen many people scale at 2200 and most that do have X3D chips for some reason. I still don't understand why there seems to be a hard cap of 2200 in most cases.

[@concinnus]

X3D chips should have much lower Infinity fabric traffic (so less stress, basically) for the same performance, because only cache misses create Infinity traffic.

[@JJFX-]

@@concinnus That makes sense and I suspect at least the dual CCD models were binned during the initial Zen 4 rollout.

[@vyathaen]

on my 7700X , performance scales even at 2200, but it will enter "crawl mode" after about 3 days testing. So I just stick to 2167, no issues at all. Ram at 6200, too much voltage for 6400.

[@felixbruckmann5702]

I set my VSOC to 1.24, tight CL28 6000 Mem timings, 2033 IF stable - but increasing to 2067, it will crash even in AIDA Benchmark...so, don't know if VSOC is too low...

[@Malklaw]

6200cl28 2200 seems stable at 1.11soc but 2200+ flck nothing even boots no matter voltages and timmings i guess ITS all cpu lottery and for me everything that boots ITS stable 😅

[@paulrichalland]

This video feels like it will attract a lot of people and that's deserved. Very interesting deep dive on how the IF works. Absolutely great content !

[@frunkaf3314]

Thank you so much for this! Ive been running 2200 FCLK thinking i got away with murder but my performance is actually better at 2167 accoridng to linpack. Im doing more testing now

[@Diapolo10]

This is the kind of content I love you for :D

[@andrexskin]

Thank you for this video, it was very interesting and don't worry about you getting carried away with the video, it's really informative

[@BaBaNaNaBa]

you can notice this instability when you get inconsistent CPU benchmark scores too

[@PaulHawke1337]

Yes! More o-scope!

[@alexmills1329]

Definitely really cool to see these errors on the scope!

[@leexgx]

I thought the infinity fabric always did this (get some slow down before crash because it was doing some ecc in the background until you pushed it to far )

[@KetogenicGuitars]

Very good tip if we need stability over anything. I'm tuning a DAW workstation where it is a thing. CPU is 7900X3D and memory 64GB of Hynix A-die. It is inside a pair of Corsair CMK64GX5M2B6000C40. It is a 6000mts module which is able to run at 6400mts. Mobo is Asrock PG x670e. I made the "easy overclock" and decided to partner it with lowering infinity fabric. At fabric 1800Mhz Passmark benchmark shows same 51000 as with fabric being 2100. I take even one of those stops is too much for audio workstation. So the retransmission is not registered by OS but action potentially stops? On top of copying buildzoid's "easy overclocking for hynix A-Die" I boosted memory voltages to 1.37 and VDD_MISC to 1.15. So Aida gives 67ns latency and 90GB memory write bandwidth. One reason to consider lower inifinity fabric is strange GPU artifacts at entering the desktop even Memtest86 passes. GPU is not in par with rest of the system but is old 2GB 1050. Another thing causing some work is mouse not being as responsive as with old 2600K system.

[@Destructor8031]

67ns isn't great, You likely have a hell of a lot of headroom to tighten up your timings. I got 6000mhz micron a die Which is literally the worst you can get and stock the latency was I over 100ns, after messing with timings it's now 65ns and rock solid stable.

[@alkestos]

@@Destructor8031 I have my 6000cl30 kit around 55ns at AIDA. @ 6000mt/s

[@mosvan]

I can't seem to get a stable FCLK above 1933 on my R7 7700x + B650M-HDV/M.2 (and high VSOC/VDD voltages). 2000 FCLK makes my cursor freeze/drop USB inputs every few seconds (checking with mousetester), and if I set FCLK>=2033 the system just refuses to post, with the DRAM and CPU post LEDs stuck on. This happens at any ram speed; I can run DDR5 6200mt/s on 1933 FCLK stable, but the clocks aren't synced 3:1, so weird latency issues occur. I don't know what the problem is between motherboard or CPU... Edit: I got a replacement 7700x and it happily runs at 2066 FCLK + 6200 memory, whereas the previous CPU had instability issues even above 1900 FCLK + 5600 memory. It's good to know that there's large variability between CPUs when it comes to the IO / infinity fabric / memory performance.

[@Ph42oN]

This reminds me of CPU-NB overclocking on phenom II. I had crappy msi motherboard where i could adjust CPU-NB clock in bios, but not voltage. So i pushed it to highest that was stable enough to boot windows. Even if with that it would run fine, but when i increased its voltage from program in windows, performance increased.

[@Overminder]

Crazy how software tools are now no longer sufficient for overclocking.

[@MiGujack3]

always had been

[@certaindeath7776]

i find it amazing, that despite the instability the cpu managed to pass all tests. for me that means for overclocks with evv i will have to do some things differently. more performance measurements for stability testing, instead of just throwing extreme workloads with validation checks on it. intersting video again.

[@georgeindestructible]

Well, since transmission error isn't the same as calculation error it makes sense that, when a successful retransmission is confirmed to have happened you should've just have performance penalty not necessarily instability, it would be extremely unstable if it didn't have some sort of error correction like retransmission, i would assume.

[@aviationodyssey1892]

Thank you for using oscilloscope! It is a great idea! 🎉

[@FreedumbHS]

For the record, the OPS in FLOPS does stand for operations per second, so your first remark about it was right

[@romangruber6685]

I had the impression Ryzen 3100 and 5800x does also the infinity error correction. And therefore you see less performance in benchmarks

[@basbass429]

Thx, awesome post. helped me to determine to really dial back from the speed since windows instantly crashes on 79503DX to estimated max fabric speed the CPU can properly run. For me it is just 1950Mhz that is for sure stable (silicon lottery lost, on this part of the chip) . 2100Mhz is instant crash in windows/windows won't even make it to boot. Will not return the CPU because on all other aspects it performs a bit better than expected boost clocks/memory timings and speed.

[@christopherjackson2157]

Now that's a helluva video thumbnail!

[@beyondspecs9401]

Thanks for the video!

[@valentin3186]

Messed up sleep schedule always provides best content AHO💯

[@simoSLJ89]

Sleep schedule is the thing I have in common with BZ. Maybe in another life we are living in US or JAP, but following the EU time zone :D

[@valentin3186]

@@simoSLJ89 heard Taiwan got some green A-die sticks for pre bin😅 Frequently catch myself in the same situation, watching this video at 5AM

[@WillFuI]

When u say best weed to test, infinity fabric stability and you say VST does that me specifically vst as like the older version or vt3? Not sure if it’s worth tracking down a download for an older version of y-cruncher to get vst

[@jerryknudsen7898]

so for those who have slower ddr5 ram in our system, would running fclock around 2000 be possible on memory 4800 cl34? I'd prefer not to have to buy more ram if possible, and i remember in ryzen 5000 and 3000 the fclock overclocking really helped close the performance gap between expensive and budget ram.

[@chapstickbomber]

I have experienced that you have to wiggle the mouse in an explorer window or on desktop, a cmd window doesn't have the same events

[@Giskarded]

Can you look into an issue with dual ccd am5 chips failing avx2 when running single core multithreaded workload at completely stock settings. No pbo no CO or expo. I have had 3 chips fail avx2 fixed occt cpu stability benchmark now. Only way to get them slightly stable or delay the errors being thrown is to give the non vcache cores positive co.

[@VicharB]

So with Ryzen 8000G series seems like AMD is defaulting to FCLK to 2400Mhz, does this mean that with latest BIOS/UEFI updates we can run Infinity at 2400Mhz as opposed to previous defaults 2000Mhz on current 7000 series? This is conjunction with DDR5 running 6400Mt/s to 8000MT/s! Currently I am running FLCK @2100Mhz (can run tests @2166 but at times unstable) with DDR 6400 CL30 on a 7700 PBO+CO optimised giving me All-Core 5.25-5.3Ghz and Single-Core 5.5Ghz.

[@Destructor8031]

No, if your bin can handle it 2200 the maximum you will get, very few people have gotten 2233. I have 2200 stable with SOC voltage if 1.11v but 2233 won't even post.

[@H53.]

Can anyone tell me what the takeaway is? What is the correct way to set everything up so the infinity fabric is stable and at peak performance?

[@atirta777]

I got a Ryzen 5 7600 build 3 days ago and been tinkering with the 2x16 5600MHz CL36 Kingston kit but can't seem to go past 5800 no matter what timings. It literally does 5800 with 28-37-37 1.4V stable with tighter timings than your easy hynix timings but blue screens at 6000 when doing anything. At 6000 it can be the loosest timings ever or CL30 + easy timings but somewhy stableness doesn't change. I get the exact same behaviour as you're getting on Infinity Fabric but it's running at 2100MHz for good measure. Is my 7600's mem controller just that bad, is this common? I don't have any other conclusions after all of this testing. Also idk if 1.25V soc voltage is too little for 6000 specifically, 1.2V works for anything at 5800.

[@ConstantinoMRL]

that must be a samsung memory

[@atirta777]

@@ConstantinoMRL Nope, it's Hynix. It says EOMH on the stick and BZ's "easy hynix timings" don't work on Samsung, hence why I could tighten them even more at 5800MHz

[@dimos15]

You should get 7600x

[@spaghettiupseti9990]

what motherboard are you using?

[@atirta777]

@@spaghettiupseti9990 B650M DS3H from Gigabyte

[@0xc0cafe8]

Can't get any performance improvements on a 7950X3D over 2033MHz, if set over 2067MHz the ram latency will be doubled in AIDA64 bench... (60->140ns+). Used to get stable on 2167MHz (~10% performance increased) on a 7600X, can't get posted if > 2167MHz. Silicon lottery.

[@PowellCat745]

Maybe single CCD CPUs can push the IF clocks higher

[@Destructor8031]

I think I might be stable at 2200 but after watching your video I'm dubious. I have a 7900X3D and when running linpack I get no errors, my mouse doesn't lag or stutter at all and I get no WHEA errors or anything. My GFLOPS varies from 577 to 584.

[@fracturedlife1393]

Nice get the oscilloscope out for the lads

[@Evenaardez]

Hey buildzoid, are your ryzen 7000 timings still good after the latest bios?

[@scottstamm7022]

This is awesome! Do Ryzen 3000 & 5000 display this same infinity fabric behavior?

[@ameykulkarni8946]

Does AM5/DDR5/7000 series/3d cache favor Windows 11 or windows 10 is still best?

[@Kage0No0Tenshi]

For intel 12th gen and amd 7000s later works best on windows 11. Windows 10 does not understand e cores of intel CPUs and amd chipset optimized really good at windows 11

[@sjoerd104]

You should only use windows 11 on intel & dual ccd ryzen chips. Windows 10 is better for single ccd ryzen chips (including 7800x3d)

[@juno1597]

Is Ryzen Master power CPU / SOC Power more accurate than HWmonitor and Hwinfo? Same for Core activity.

[@dracleirbag5838]

Can you give us a process to optimize the system you said memory clock then infinity fabric what about the main CPU clock

[@gamingfromjohnwayne]

What do u get from adjusting this.i have ryzen 7 7700 non x

[@cooper23231]

What program can i use to stress test my mem for free? I'm crashing in a specific game after tunning it the way he showed in an earlier vid.

[@Rip-Van-Tinkle]

Very interesting, thanks for this 👍

[@madd5]

I think they've disabled it with later bios. Because I couldn't run 2133mhz before. Now I can run 2167mhz. Previously only 2067mhz was stable on my 7950x.

[@Jarrahfe]

my tCL, tRCDWR and tRCDRD is is 32 38 38 / while RAM labeled as 30 40 40 - how to fix this ? already 30 38 38 edited on bios, but on windows it's showing 32 38 38... how to fix this ?

[@erickelly4107]

My R7 7800X3D ASUS ROG STRIX X670E-F / 1709 2X 16GB 6200MT/s 28-36-32-30 (tight subs) 2200 FCLK. 57.8 NS Aida64. PYPrime: 7.554 sec. VDD: 1.45v VDDQ/IO: 1.30v VSOC: 1.24v VDDG CCD: 1.050v VDDG IOD: 0.950v Linpack Xtreme 10GB 50 loops stress test: Avg: 405.28 GFlops. (Better than FCLK lower than 2200) Every pass is ~ 70.5 seconds. At FCLK 2233 system boots but that’s it, can barely move mouse and system resets within seconds. At FCLK 2200 apparently perfectly stable but totally useless at 2233 FCLK. I would have thought 2233 FCLK would have been at least “somewhat stable” but nope it’s a hard no go. Tried 3 different times just to be sure.

[@MarkoBelic-xq5ce]

Can i see your zen timings?

[@erickelly4107]

@@MarkoBelic-xq5ce Unfortunately can’t post Zen Timings here on KZbin but I’ve since improved Latency down to 57.8 ns and PYPrime to 7.554s. Primaries slightly tighter: 6200MT/s 28 36 32 30 and tRFC: 372 (120ns)

[@Destructor8031]

Yeah I got 2200 rock solid stable but 2233 won't even post lol

[@VetteTTV12]

Can the CPU be damaged from an unstable IF? I had what I thought was a stable memory OC with 1.25v on SOC, VDDQ, VDDIO ; 1.4v VDD and IF @ 2100mhz. It passed Linpack 4hr run and Y-cruncher for 4hrs. Buy after running a bunch of benchmark, like 3dmark and other similar programs, the system has become very unstable. Trying to run Prime95 results in an instant lock up. I reset CMOS and did memtest86. Memtest86 came back clean. Ran Prime95 for 8 hours on stock settings and everything was fine. But if I try to tweak the memory AT ALL the system is instantly unstable. System is x670e Aorus Xtreme, 7950X, 7900XTX, and Corsair dominator running Hynix A-die. Any advice? Could the CPU be damaged or the motherboard having issues? I'm still within the return window for the these parts, buy only for another week... honestly regretting going AMD at this point.

[@paradoxicalcat7173]

1.25 V on VDDIO? I understand stock is 0.95 V. You probably killed it.

[@alphaomega1969]

Increasing Infinity cache alow you to tell the cpu the speed of you ram so all its core can access it.

[@alphaomega1969]

AM3 motherboard goes up to 2600mgz on cpu northbridge also known as memory controller on cpu and known as Infinity Cache today. This is the speed of motherboard and cpu single and dual channel 😎

[@areng871]

Then cezzane is still the king on this matter. It runs the fabric 2500 with no problem on mine

[@Moon-ty2hn]

that's because it's a monolithic design

[@previousslayer2]

yea at dumb volts, i've had my 5700g at 2500 1.5 vsoc for an eternity but we'll see in another year how it goes xd

[@ktg_2castle960]

Is this the fix for AMD dip? I’m looking at going intel rn for higher lows

[@crispysilicon]

Nothing like struggling silently until death. Bravo AMD. Really is impressive how long it deals with it sometimes.

[@SaccoBelmonte]

When you push the IF far enough it creates a Hardcore band logo.

[@mho...]

isnt a glitch in the infinity fabric basically what a stroke would be for a human?! sounds serious!, having a drop like that in the mix 😨

[@user78405]

Higher the memory clock, the higher fabric clock gets equals more voltage for soc, then burn chips begin without users knowing

[@bgtubber]

Fabric clock and memory clock are decoupled in Ryzen 7000. You can tweak them independently and you won't have performance degradation like in previous gen Ryzens if they are not in sync.

[@TheHalo294]

amazing content again :D

[@Zerf7]

This helps with what? Cpu oc limits?

[@ConnorDaly-n7c]

You know a lot about this stuff so help a newbie out, will you? Yeah, it's a really novice question but frankly, I'm clueless despite my hours of research. For the i5 13600k processor, the officially supported DDR5 memory speed is 5600 MT/s which is equivalent to 5600 MHz (I THINK). I know full-well it can run faster memory speeds - 6000, 6400, 6600, maybe more - with a basic (XMP profile in BIOS) or fine-tuned manual overclock. I also know that with Intel, you want to prioritize speed and not CL. With that in mind, will native 5600 MHz D5 RAM provide any stability benefits with the 13600k over native 6000+ MHz RAM? Or is it best to get 6000+ MHz RAM and just immediately enable a basic XMP profile for the CPU? If it helps I plan on undervolting the 13600k too.

[@ConnorDaly-n7c]

mm I just realised that the default run speed for RAM (that I was interested in at least) is 4800 MHz so with that in mind, I don't think it matters and it's always worth getting the RAM with the higher MHz rate so in this case, 6600. I'm still not sure what the timings would be if you got 6400 MHz RAM and ran it lower than that but with Intel it's not as big of a problem. I imagine they would be tighter...

[@K9PT]

Confirme please that in 30/06/2024 this is not the same limite right?!

[@Reformedekko]

Pushing ryzen 7000's fabric to infinity?

[@otbricki]

and beyond

[@redrock425]

​@@otbricki Lol I knew your reply would say that.

[@niks660097]

IF clock speeds shouldn't matter, its Bytes/clock bandwidth, latency and network topology, even if IF were to be 1GHZ and it could still power 100s of pcie 5 lanes and 10 GHZ ddr5 without sweat, if next gen IF has some genius topology like fujistsu A64FX it could kill all monolith designs..

[@JohannesDahl42]

According to Wikipedia, FLOPS includes "per second", so saying "GigaFLOPS per second" is definitely redundant. I suppose an argument could be made that you were using the plural form of FLOP, thus FLOPs, but that's confusing, and we don't do that with other measures: "2 kilometres" -> "2 km", not "2 kms".

[@ertai222]

Ive been at 2200 for about the last month or so with my ram at 6400 on my 7800x3d. Seems stable enough to me🤷🏻‍♂️ It seems to be the only setting that gives me the most performance. Going lower to like 2167/2133 etc actually gives me less performance. I lose read/write speeds in Aida64 with anything less than 2200. I've played through Jedi Survivor with no issues other than the game crashing because it's a buggy mess. The fact ive been running these settings with no apparent issues means i will continue to do so for the foreseeable future.

[@IT10T]

That's the spirit; if it ain't broke, don't fix it.

[@Underground.Rabbit]

That's a 3% difference. That's a placebo youre feeling.

[@jonaaa2036]

I'm running at 2200 as well, 7800X3D.

[@PowellCat745]

I’m wondering if the 7800X3D has a different substrate that can offer higher infinity fabric clocks. The number of CCDs may also have something to do with infinity fabric clocks.

[@Malklaw]

Same for me 7600x anything lower than 2200 3k lower Aida scores so ITS kind of turbo button 😂

[@Bojcha76]

i wonder how's doing 1ccd cpu.

[@mrjoerizk]

You are awesome!

[@xedsu]

could I pay you to optimize my bios for gaming

[@aliensounddigital8729]

Video from the future eh?

[@eugene1640]

Thank you

[@DannyGodhouse]

Cool wallpaper! LOL

[@troeteimarsch]

Does the same apply for Intel cache / uncore ? I'm at 4800 and have no clue how to test its' stability.

[@ActuallyHardcoreOverclocking]

no. If you crash the ring on an intel CPU the entire system hard locks and even the reset button stops working so you have to hold the power button for 10 seconds to force a shutdown.

[@troeteimarsch]

@@ActuallyHardcoreOverclocking thx. How can I try to crash it?

[@RaukGorth]

So the 2033 recommendation still stands?

[@WoNCrawler]

That's why i recommended everyone to upgrade their AM4 stuff instead of using AM5 boards... the first 1-2 gens in tech are always public bug tests.

[@Ironclad17]

AM5 has some pretty glaring issues but this isn't so much a bug as a newer feature that isn't well known. Previously IF OC was either stable or completely failed under load or boot.

[@PowellCat745]

Is it even worth it to OC your infinity fabric if you only run your RAMs at 6000 MT/s?

[@ActuallyHardcoreOverclocking]

if it's stable it can lower your latency a bit.

[@PowellCat745]

@@ActuallyHardcoreOverclocking Thank you! It does look like my latency went down about 2ns with IF == 2100. Curiously, my north bridge clock will halve to 933 when I set my memory speed to 6200. Didn’t seem to affect performance tho.

[@optimiz4]

Rather than error correction, could this be a bug in the driver layer where Infinity Fabric errors are not getting surfaced to the OS on 7000 series CPUs?

[@ActuallyHardcoreOverclocking]

without ECC the system should just crash.

[@stanimir4197]

@@ActuallyHardcoreOverclocking ECC is generally free - if it's correction indeed. The way you explain it - it's more like error detection and retransmition. The ECC requires extra bits per word (e.g. 1 extra bit per 8 bits), detection can be implemented just with checksum, which is a lot cheaper.

[@nivea878]

AMD is a PTSD monster

[@snajper2222]

7800x3d + 6000mhz ram What is stable and programs do not detect?

[@jamisonw.327]

What is the memory manufacturer? Do they use Samsung, micron, sk hynix? Which die style? B die? Etc.

[@snajper2222]

@@jamisonw.327 HWiNFO64: Ram: Corsair VENGEANCE CMK32GX5M2D6000Z36 - Samsung B-Die (4.2) Year: 2023 Week: 4 Motherboard: gigabyte x670 aorus elite ax rev 1.1 Bios: F8 Processor: AMD Ryzen 7 7800X3D

[@GewelReal]

I wonder if AMD will update the IO die. DDR5 6000 is pretty slow

[@Pro4TLZZ]

Thanks

[@isaac3045x]

amdip?

[@giantnanomachine]

I can't get 6200 stable with 4x 16GB Hynix-M. Is it worth continuing to try or is 6200 just not in the cards with 4 DIMMs?

[@snaky115]

6200 is a struggle even with 2 dimms half the time, my 7800x3d cannot do 6200 even with JEDEC loose, like 50-50-50.. Though feels like my 7800x3d IMC is a giga-potato in general, as doing tighter-than-XMP at 6000 is already a struggle.

[@Heisenberg355]

Everything just tanks performance instead of crashing :( FCLK, clock stretching, VRAM....

[@jpteknoman]

mine does that at 2000 even without running a test

[@Durayne]

I currently try to get my 7800x3d stable with 2x32 6000 cl30. Everything is fine ... every Benchmark ... fine. Until I launch Battlefield 5. That causes hard resets, like literally power off and restart once expo is activated.

[@mircomputers]

i found that Geekbench 6 also was crashing mine while corecycler was fine for hours

[@Durayne]

Yeah currently trying to figure out whether its the infinity fabric or the ram thats causing the instabilities or a combination. Maybe Ill test y-cruncher since buildzoid mentioned it stresses the infinity fabric.

[@IvanUskov]

I had similar issues with BF5 by the too strong CPU undervolting reason. All other tests also were fine. My memory is 2x32 6000 cl 32, no any XMP/EXPO activation, manual settings.

[@ShadowbannedAccount]

It's better to stress test than to use benchmark.

[@Durayne]

@@IvanUskov Might test that again. For me, I think I tried Curve Optimizer stepping lower. And only expo, but maybe I still test.

[@unsivilaudio]

This is damn cool to watch though.

[@johngalt8708]

Is there a guide on how we can check this on our own overclocks to see this performance issue?

[@alphaomega1969]

Ram speed should not go above the infinity cache or you will bottle neck your system and have crash issue

[@Taraquin83]

Interesting!

[@zuckdaddy1596]

Neat!

[@johnpaulbacon8320]

Hello All

[@yoochoobb]

*its :)

[@hughJ]

This will be how we distinguish AI from real people in the future. It's like the inverse of Data being unable to use contractions (only AI will use them correctly.)

[@emperorSbraz]

mega interesting

[@KRAVER_]

AMD :)